FutureSearch – Benchmark for Language Model Forecasting

[Navigating Transformative AI] Open Philanthropy recommended a grant of $606,600 to FutureSearch to support the development of a benchmark to evaluate the ability of large language models (LLMs) to forecast geopolitical events. The benchmark will be freely available to the public. This grant was funded via a request for proposals for projects benchmarking LLM agents on consequential real-world tasks. This falls within our focus area of potential risks from advanced artificial intelligence.